Research on dirt road autonomous driving technology is increasing as a way to improve the operability of the fixed-type surveillance and reconnaissance systems by mounting them on a movable platform such as unmanned vehicles. Environmental awareness technology such as environment recognition and traversable region analysis is one of the important research topics for dirt road autonomous driving.
In this paper, we propose the 3D LIDAR-based traversable region analysis methods for autonomous driving of unmanned vehicles for surveillance and reconnaissance operating on unpaved roads. Two approaches are used for detection of traversable regions by analyzing the 3D LIDAR data obtained from the dirt road. One is using the geometric characteristics by defining the lateral slope of the landform around the road. The other is taking advantage of the reflectivity of 3D LIDAR as a physical feature to enhance a discriminating capability of the material on the ground near the traversable area.
A model for the travelable area is created based on the identified features obtained from the pre-analysis information of the traversable area. The sensor data of the unit frame is then transformed into a probability map of the movable area using the generated movable area model, and the map of the travelable area that is sequentially generated is updated with reliable maps through Bayes formula based probabilistic updating.
The proposed method was applied to a prototype unmanned vehicle of surveillance and reconnaissance, and autonomous driving tests were carried out in the environment of dirt road. Using the probabilistic map generated by the proposed method, autonomous driving tests were successfully performed at an average speed of 20km/h and a maximum speed of above 35km/h at the off-road environment.